Implementing the Morphology-Syntax Interface: Challenges from Murrinh-Patha Verbs
Total Page:16
File Type:pdf, Size:1020Kb
IMPLEMENTING THE MORPHOLOGY-SYNTAX INTERFACE: CHALLENGES FROM MURRINH-PATHA VERBS Melanie Seiss Universitat¨ Konstanz Proceedings of the LFG11 Conference Miriam Butt and Tracy Holloway King (Editors) 2011 CSLI Publications http://csli-publications.stanford.edu/ Abstract Polysynthetic languages pose special challenges for the morphology- syntax interface because information otherwise associated with words, phrases and clauses is encoded in a single morphological word. In this pa- per, I am concerned with the implementation of the verbal structure of the polysynthetic language Murrinh-Patha and the questions this raises for the morphology-syntax interface. 1 Introduction The interface between morphology and syntax has been a matter of great de- bate, both for theoretical linguistics and for grammar implementation (see, e.g. the discussions in Sadler and Spencer 2004). Polysynthetic languages pose special challenges for this interface because information otherwise associated with words, phrases and clauses is encoded in a single morphological word. In this paper, I am concerned with the implementation of the verbal structure of the polysynthetic language Murrinh-Patha and the questions this raises for the morphology-syntax interface. The Murrinh-Patha grammar is implemented with the grammar development platform XLE (Crouch et al. 2011) and uses an XFST finite state morphology (Beesley and Karttunen 2003). As Frank and Zaenen (2004) point out, a morphol- ogy module like this in combination with sublexical rules makes a lexicon with fully inflected forms unnecessary, which is especially important for a polysyn- thetic language as listing all possible morphological words would be unfeasible, if not impossible. However, this raises the question of the division of work be- tween syntactic grammar rules in XLE and morphological formations in XFST. By looking at different cases of long distance dependencies within the Murrinh-Patha verbal template, this paper contributes to an understanding of the division of work between these two components of grammar. The paper is structured as follows. Section 2 gives a short overview over the Murrinh-Patha verb and provides some examples of the complexities of the verbal template. Section 3 then outlines the assumed architecture for the morphology- syntax interface. The first part of the section summarizes the theoretical framework put forth by Butt and Kaplan (2002). This framework is then illustrated by the basic outline of the Murrinh-Patha XLE and XFST implementation. Sections 4 and 5 discuss the details of the implementation of the dependen- cies in the verbal template. Section 4 deals with the implementation of dependen- cies concerning subject and object number marking as well as tense marking. I yI would like to thank Rachel Nordlinger who kindly provided me with field data and information about the language. Further thanks go to my supervisor Miriam Butt and the audience of the LFG Conference 2011 for discussions about this topic, and to Jane Simpson and the members of the Language Technology Group at Macquarie University for discussions on previous versions of the talk. argue that these dependencies should be considered morphological dependencies and should thus be modeled within the morphological component. In contrast to the purely morphological features discussed in section 4, sec- tion 5 introduces dependencies between two lexical-semantic components of the Murrinh-Patha verb. Although the combination of these components is also part of the word formation process, I argue that the dependencies are nevertheless more efficiently modeled in the syntactic component, as their combinatory possibilities depend on syntactic features. Dependencies within a word thus do not have to be modeled in the morphology exclusively. Especially in polysynthetic languages some dependencies may also involve syntactic features and can more efficiently be implemented in the syntax. 2 Dependencies in the Murrinh-Patha Verbal Template Murrinh-Patha is a non-Pama-Nyungan language spoken around Wadeye in the Northern Territory of Australia. It is a headmarking, polysynthetic language with a very complex verbal template and minimal case morphology on the noun. Due to space limitations only some examples of the dependencies within the verbal template can be discussed here. For a more detailed overview of the complexities of the verbal template see Nordlinger (2010b), and see Blythe (2009) for a more general introduction to the language. The Murrinh-Patha verbal template can be considered to have nine different slots for verbal stems, agreement markers and incorporated body parts, adverbials or particles. This paper mainly discusses the slots 1, 2, 5, 6 and 8 and their inter- dependencies. 1 2 3 4 5 6 7 8 9 Class. SubjN/ RR IBP Lex TNS Adv/Prt SubjN/ Adv/Prt Obj APPL ObjN Class: classifier stem, marked for tense, aspect & subject number SubjN: subject number markers for dual & paucal subject Obj: object agreement marker ObjN: object number marker for dual & paucal RR: reflexive / reciprocal marker IBP: incorporated body part APPL: applicative marker -ma Lex: lexical stem TNS: tense marker Adv/Prt: Adverbial / Particle Figure 1: Murrinh-Patha verbal template (adapted from Blythe 2009) The semantics of the Murrinh-Patha verb is determined by two morphemes to- gether, the so-called classier stem and the lexical stem. There are 38 different classifier stems. Classifier stems inflect for tense, aspect and subject number and they encode this information in portmanteau forms. They mostly have a generic meaning. In contrast, lexical stems may express more specific meaning and are also noninflecting. (1) shows some first simple examples.1 In (1a), the classifier stem SNATCH(9) combines with the lexical stem rta, ‘hug’, while in (1b) the classifier stem 13 com- bines with the lexical stem ngkardu ‘see’. (1c,d) show that not every lexical stem can combine with every classifier stem and vice versa. The details of these com- binatory possibilities and their consequences for the implementation will be dis- cussed in section 5. (1) a. manganta mangan-rta 3sgS.SNATCH(9).nFut-hug ‘He/she hugged him/her.’ (Nordlinger, 2010a) b. bamkardu bam-ngkardu 3sgS.13.nFut-see ‘He/she saw him/her.’ (Fieldnotes R. Nordlinger) c. *mangan - ngkardu d. *bam - rta The classifier stems are inflected for tense. However, in all but the non-future tense, a corresponding tense marker has to attach to the classifier stem in slot 6. For example, the tense marker -nu has to attach to the future classifier stem form, as in (2a). In (2b), the tense marker -dha attaches to the past imperfective form of the classifier stem. (2c) shows that only corresponding tense markers can attach to the classifier stems. (2) a. ba - ngkardu - nu 1sgS.SEE(13).Fut - see - Fut b. be - ngkardu - dha 1sgS.SEE(13).PImpf - see - PImpf c. *ba - ngkardu - dha 1sgS.SEE(13).Fut - see - PImpf While the tense dependencies can be considered as simple agreement, the rela- tionship between the number information encoded in the classifier stem and the 1Because of their generic meaning, classifier stems are glossed either with capitals or a number; lexical stems are glossed according to their meaning in combination with the classifier stem. Ab- breviations used in the glosses are: RDP: reduplicated; Fut: Future tense; PImpf: Past Imperfective; du.f: dual female subject; pauc.f: paucal female subject; 1sgDO: 1st singular direct object marker; etc. separate subject number markers in slots 2 and 8 is more complex. The classifier stems themselves show a three-way number contrast: singular, dual and plural. The singular classifier stem can combine with a separate dual number marker which overwrites the information of the classifier stem, as can be seen in (3a). The dual classifier stem without a separate subject number marker denotes the dual, non- sibling category as in (3b). However, this information can also be overwritten by attaching a paucal number marker as in (3c). (3) a. bam - ngintha - ngkardu 3sgS.SEE(13).nFut - du.f - see ‘They two (non-siblings) saw him/her.’ b. pubamka - ngkardu 3duS.SEE(13).nFut - see ‘They two (siblings) saw him/her.’ c. pubamka - ngkardu - ngeme 3duS.SEE(13).nFut - see - pauc.f ‘They few (non-siblings) saw him/her.’ As can already be seen in these simple examples, the number markers for dual and paucal subject behave differently in the verbal template. The paucal number marker attaches after the lexical stem (in slot 8), while the dual number marker attaches before the lexical stem (in slot 2). However, the placement of the dual number marker is more complex as can be seen in the examples in (4). The subject number marker competes with the object markers for slot 2. Thus, if an overt object marker is present, the subject marker has to move to slot number 8 ((4a)). However, as can be seen in (4b), if no overt object marker is present, the subject number marker is disallowed in slot 8. (4) a. bam - ngi - ngkardu - ngintha 3sgS.SEE(13).nFut - 1sgDO - see - du.f ‘They two (non-siblings) saw me.’ b. *bam - ngkardu - ngintha 3sgS.SEE(13).nFut - see - du.f These quite simple examples already show that the Murrinh-Patha verbal template is quite complex. It involves long distance dependencies as well as constraints which depend solely on the linear ordering within the template, and are not con- strained by the functions the morphemes fulfill. Moreover, even more complex cases exist, e.g. with discontinuous object markers, in which case the subject num- ber markers cannot be expressed overtly and the form is ambiguous. These ex- amples, however, are too complex to be treated in detail here and the simpler ex- amples in (1)-(4) suffice to illustrate the analysis. Before going into the details of the analysis, though, the next section lays out the general architecture of the morphology-syntax interface assumed in this paper. 3 Morphology - Syntax Architecture As morphemes in polysynthetic languages often fulfill similar functions as words fulfill in non-polysythetic language, the dependencies described in the previous section could in principle either be modeled in the syntax or in the morphology.